On September 29, 2026, OpenAI launched GPT-6.1 Sol at DevDay 2026. The pitch is simple: this model nearly matches GPT-6 Astra on coding, computer use, and professional work, and it costs roughly one-fifth as much to run. If you have been holding off on building AI-powered workflows because Astra pricing made the math ugly, the math just changed.

This is not a minor iteration. GPT-6.1 Sol beats its predecessor, GPT-6 Sol, by 6.4 percentage points on DeepSWE v1.1 — a benchmark that tests real software engineering tasks in actual codebases — while costing less per call. On AutomationBench, which measures multi-step business workflows across 47 tools in sales, marketing, operations, support, finance, and HR, it scores 2.2 percentage points above Claude Opus 5.5 at medium reasoning effort, at roughly a third of the cost. On Terminal-Bench Science, it more than doubles GPT-6 Sol's score at maximum reasoning effort while costing less than half per task.

The pricing that actually matters

The headline numbers from OpenAI's announcement are worth repeating because they are the reason this model exists. Standard input and output tokens are priced at roughly one-fifth of Astra's rates. Cached input — the tokens you reuse across requests — costs $0.10 per million tokens. That is 95% less than standard input pricing and 50% less than GPT-6 Sol's cached input rate.

Why does cached input matter? Because most AI-powered business workflows are repetitive. You send the same system prompt, the same tool definitions, the same reference documents, over and over. The only thing that changes is the user's query. Caching lets you pay once for the static parts and reuse them at a fraction of the cost. If your workflow runs 1,000 times a day with a 4,000-token system prompt, caching saves you real money every month.

The practical guide OpenAI published alongside the launch puts it bluntly: "Run independent tasks together so one slow step doesn't hold up unrelated work." That is the production playbook. Use caching for recurring context. Use compaction to shrink long conversations. Measure cost per successful task, not cost per token. The model is cheap enough now that the optimization work pays off faster.

What it means for small business AI workflows

If you are running a WordPress site and using AI to classify support tickets, generate content drafts, or route leads, GPT-6.1 Sol is the model that makes the unit economics work. A support ticket triage workflow that classifies and routes 500 tickets per day costs significantly less on Sol than it did on Astra, and the quality difference is small enough that most users will not notice.

If you are building with n8n, Make, or Zapier, the practical impact is even more direct. n8n's AI Agent node, Make's AI Agents module, and Zapier's AI Actions all let you specify which model to use. Switching from Astra to Sol for everyday tasks — lead scoring, email summarization, data extraction, content generation — cuts your AI spend by roughly 80% without a meaningful drop in output quality.

The benchmark that matters most here is AutomationBench. If GPT-6.1 Sol can reliably complete multi-step business workflows across 47 tools, it means you can build automations that make decisions, call APIs, write to databases, and follow up — all on a model that does not require a CFO's approval to run.

The DevDay announcements behind the model

GPT-6.1 Sol did not ship alone. OpenAI used DevDay 2026 to announce more than 20 updates, including the opening of ChatGPT as a shared surface where humans and AI agents can collaborate. The company now reports 1.2 billion weekly users across ChatGPT and its API surfaces. That number is relevant because it means the ecosystem of tools, integrations, and templates built around OpenAI's models is not going anywhere.

The practical guide also introduced compaction — a feature that shrinks long-running conversations while preserving the state needed to continue. For anyone building multi-step agents that maintain context across dozens of turns, compaction solves the problem of context windows filling up and costs spiraling. Combined with caching, it means an agent that runs for 50 turns costs a fraction of what it did six months ago.

When you still need Astra

GPT-6.1 Sol is not a replacement for Astra on every task. OpenAI's own benchmark data makes this clear. On Terminal-Bench Science at maximum reasoning effort, Astra scores 68.1% — the highest of any tested model — while Sol is competitive but not at that level. For the hardest scientific research, complex multi-agent coordination, or tasks that require the absolute best reasoning under pressure, Astra is still the right pick.

For everything else — coding, document analysis, business automation, computer use, content generation — Sol is now the default choice. The price gap is too large to justify running Astra on tasks where Sol delivers comparable results.

What to do right now

If you are already using GPT-6 Astra in production workflows, audit your usage. Look at every API call and ask whether the task actually requires Astra-level intelligence. Classification, summarization, data extraction, and routine code generation almost certainly do not. Switch those calls to GPT-6.1 Sol and measure the results for a week. If the quality holds — and the benchmarks suggest it will — you just cut your AI bill by 80%.

If you are building new workflows, start with Sol. The caching economics make it viable for high-volume use cases that were not cost-effective before. A lead-scoring pipeline that runs 10,000 times a day, a content moderation system that reviews every user submission, a support bot that handles first-response on every ticket — these are now affordable at small-business scale.

If you have not started building with AI at all, this is the model to start with. The pricing is low enough that experimentation is cheap, the capability is high enough that production use is realistic, and the ecosystem is mature enough that you will not be building alone.

The gap between "AI is too expensive for my business" and "AI is a tool I use every day" just got a lot smaller. GPT-6.1 Sol is the reason.

Sources: OpenAI — Introducing GPT-6.1 Sol, OpenAI — A Practical Guide to Building with GPT-6, OpenAI — DevDay 2026 Recap